Thread: Meson add host_system to PG_VERSION_STR

On Wed, Nov 16, 2022 at 12:08:56AM +0100, Juan José Santamaría Flecha wrote:
> As mentioned here [1] it might be interesting to complete the returned
> information by version() when compiled with meson by including the
> host_system.

The meson build provides extra_version, which would be able to do the
same, no?  The information would be appended to PG_VERSION_STR through
PG_VERSION.
--
Michael

On 16.11.22 01:01, Michael Paquier wrote:
> On Wed, Nov 16, 2022 at 12:08:56AM +0100, Juan José Santamaría Flecha wrote:
>> As mentioned here [1] it might be interesting to complete the returned
>> information by version() when compiled with meson by including the
>> host_system.
> 
> The meson build provides extra_version, which would be able to do the
> same, no?  The information would be appended to PG_VERSION_STR through
> PG_VERSION.

I think this is meant to achieve parity between the version strings 
generated by configure and by meson.

Perhaps some examples before and after on different platforms could be 
shown.

Hi,

On 2022-11-16 09:01:04 +0900, Michael Paquier wrote:
> On Wed, Nov 16, 2022 at 12:08:56AM +0100, Juan José Santamaría Flecha wrote:
> > As mentioned here [1] it might be interesting to complete the returned
> > information by version() when compiled with meson by including the
> > host_system.
>
> The meson build provides extra_version, which would be able to do the
> same, no?  The information would be appended to PG_VERSION_STR through
> PG_VERSION.

I don't really follow: Including the operating system in PG_VERSION_STR,
as we're doing in autoconf, seems orthogonal to extra_version? Adding linux
into extra_version would result in linux showing up in e.g.
SHOW server_version;
which doesn't seem right.


I think there's a further deficiency in the PG_VERSION_STR the meson build
generates - we use the build system's CPU. Autoconf shows $host, not $build.


For comparison, on my machine autoconf shows:
  PostgreSQL 16devel on x86_64-pc-linux-gnu, compiled by gcc-12 (Debian 12.2.0-9) 12.2.0, 64-bit
whereas with meson we currently end up with
  PostgreSQL 16devel on x86_64, compiled by gcc-13.0.0

I still don't think it makes sense to try to copy (or invoke)
config.guess. Particularly when targetting windows, but even just having to
keep updating config.guess in perpituity seems unnecessary.

Given we're looking at improving this, should we also add 32/64-bit piece?

If so, we probably should move building PG_VERSION_STR to later so we can use
SIZEOF_VOID_P - configure.ac does that too.

With extra_version set to -andres the attached results in:

PostgreSQL 16devel-andres on x86_64-linux, compiled by gcc-13.0.0, 64-bit

Greetings,

Andres Freund

On Wed, Nov 16, 2022 at 02:12:05PM +0100, Juan José Santamaría Flecha wrote:
> On Wed, Nov 16, 2022 at 10:50 AM Peter Eisentraut <
> peter.eisentraut@enterprisedb.com> wrote:
>> Perhaps some examples before and after on different platforms could be
>> shown.
>
> Yes, that would make clear what the patch is trying to do. For version() we
> get:
>
> Configure:
>  PostgreSQL 16devel on x86_64-pc-linux-gnu, compiled by gcc (Debian
> 6.3.0-18+deb9u1) 6.3.0 20170516, 64-bit
>
> Meson:
>  PostgreSQL 16devel on x86_64, compiled by gcc-6.3.0
>
> Patched:
>  PostgreSQL 16devel on x86_64-linux, compiled by gcc-6.3.0

Ah, thanks.  I was not following this point.  Adding the host
information for consistency makes sense, indeed.
--
Michael

Hi,

On 2022-12-09 14:53:10 +0100, Juan José Santamaría Flecha wrote:
> I've added an entry [1] in the commitfest so we don't miss this subject.

I indeed had forgotten. Pushed now.

Greetings,

Andres Freund